Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog.

[2.0.0] - 2026-06-22

API Key hashing now uses SHA-256 + pepper (SECRET_KEY) with timing-safe comparison via secrets.compare_digest()
CORS validation rejects ["*"] when allow_credentials=True (Pydantic model_validator)
Scope validation: ClientCreate.scopes rejects invalid values via field_validator
Master key comparison uses constant-time compare

Dead Letter Queue (DLQ): Failed ARQ jobs pushed to Redis DLQ after PIPELINE_MAX_RETRIES attempts. Admin endpoints to list/retry/clear dead jobs (GET/POST/DELETE /admin/api/queue/dead)
Circuit Breaker AI: All AI calls wrapped with asyncio.wait_for() + exponential backoff retry (configurable via AI_REQUEST_TIMEOUT, AI_MAX_RETRIES)
Request ID Middleware: X-Request-ID header injected in all responses + bound to structlog context
Rate Limiting per Plan: Dynamic rate limits based on client plan (free/standard/premium) via ContextVar
Redis Rate Limit Storage: Optional persistence via RATE_LIMIT_STORAGE_URL
Worker script: worker.py for standalone ARQ worker launch
Docker Compose stack: PostgreSQL 16 + Redis 7 + MinIO + API + Worker
.env.example: All configuration variables documented
Grafana Dashboard: docs/grafana-dashboard.json with 10 panels (images, tokens, pipeline, storage, WebSocket, errors)
Database indexes: Composite indexes (client_id, uploaded_at) and (client_id, processing_status) on images table

APP_VERSION aligned to 2.0.0 in config (was 1.0.0)
delete_files() now fully async — uses await backend.delete() instead of fire-and-forget ensure_future()
Upload flow: files deleted on DB commit failure (no orphaned S3 files)
Gemini client: cached with TTL — recreated if GEMINI_API_KEY changes
parse_redis_url() extracted to app/workers/redis_client.py (DRY — was duplicated in main.py and image_worker.py)
Rate limit key function encodes plan for per-plan bucket isolation
_FallbackArqPool moved to app/workers/arq_fallback.py (avoids circular imports)
ARQ worker: configurable PIPELINE_TIMEOUT and PIPELINE_MAX_RETRIES, on_shutdown handler
Pipeline: push_dead_job() on final retry failure marks image as ERROR with DLQ metadata
Admin panel: VITE_API_BASE_URL changed from internal Docker hostname to http://localhost:8000

ARQ fallback now returns explicit warning in UploadResponse.message instead of silently ignoring
Removed obsolete comment in ai_vision.py about StorageBackend.read
Removed signal handler instability in test fixtures
Fixed Docker Compose admin connectivity (browser couldn't resolve internal backend hostname)

WebSockets / Real-Time:
- Subscriptions on per-image workflows (/ws/pipeline/{image_id}).
- Redis 60-second Event Buffering for reconnections.
- Admin Monitor WebSocket feed (/ws/admin/monitor).
REST API Versioning:
- Global /api/v1/ prefix for structural stability.
- Added strict API Sunset Date tracking middleware.
Python SDK (imago-client):
- Publishable PyPI-ready package located inside sdk/.
- Automatic httpx HTTP reconnects, WebSockets streaming natively wrapped, and robust generic Error classes (ImagoError, AuthError, etc.).
Admin Backend APIs:
- Complete JSON statistics (/admin/api/stats), global client configurations (/admin/api/clients/*) and Queue observability natively piped to the existing React Interface.
Shaarli Integrations:
- Skeleton integration middleware (integration/shaarli) capable of bridging Shaarli's events right into Imago using the official SDK.
Extensive Documentation:
- Modular guides provided inside docs/ detailing WebSockets, Integrations, and SDK consumption.

All API routing endpoints have been moved behind the /api/v1 namespace. Existing API Clients must update base URLs from / to /api/v1/.

Fixed unawaited coroutines in async mocks causing TypeError on Pytest environments.
Fixed an incorrect payload key mapping in APIClient responses causing 500s (AttributeError).
Mocked S3 and ARQ correctly globally to ensure local automated tests pass completely isolated from network logic.